Archdaily - How does AI perceive the contemporary home?
To think about how we inhabit is to think about architecture. If the primal need for shelter gave rise to the discipline, today, housing remains one of architects' most significant concerns. Providing comfort, seeking innovative materials, respecting memory, transforming culture - multiple layers intersect in a residence's design. Therefore, imagining the synthesis of the contemporary home is a great challenge. In search of new perspectives, we collaborated with Archdaily to understand how Artificial Intelligence (AI) perceives the contemporary home in the context of 15 different countries. Among data that touches on facts of reality and fiction, the pictures that emerge can bring inspiration and important reflections on spatial practice and the creation of its images.
Working with AI software that generates images based on commands, such as Midjourney or DALL-E, allows for the creation of distinct approaches that arise from the mathematical data of each tool. Designing a custom dataset, inputting precise coordinates, and editing or adding elements in post-production is essential to the outcome of an image. With this in mind, we embarked on this project with the question: how does artificial intelligence perceive the contemporary home?
Without a clear rule, the first step of the research was to input a prompt into Midjourney with the following words: "/imagine: contemporary house showcased on archdaily website." As a result, images with very similar language and few variations in terms of materials and context emerged. This homogenization occurs due to the predominant architectural styles in the database, from which the software derives a pattern through identification or classifications that are created during its training. For example, the dataset of images may perceive that most residences published on ArchDaily have two floors, orthogonal lines, with communal functions on the ground floor and a more private zone on the upper level with large glass openings. These standard features generate the pattern by which the software conforms to the requested image. Excluding other architectural approaches, techniques, and materials that are possible for a contemporary home.
In this way, artificial intelligence, through its machine learning, tends to perpetuate a hegemonic language of architecture and does not consider other possibilities, such as ancestral or vernacular architectures, or those that do not follow what is produced on a large scale. There is thus a repetition given by automated decisions that can directly reflect how we perceive the world, influencing the way we imagine and design architecture. As noted by Giselle Beiguelman, a professor at Faculty of Architecture and Urbanism at the University of São Paulo (FAUUSP), "no technology is neutral, and every photograph is a product of a series of technical and programmatic instances that range from framing modes to what is left out of the scene." An essential statement for architects to effectively manipulate AI tools, maintain innovation in their processes, and propose appropriate solutions for different contexts.
In search of a broader approach to how artificial intelligence perceives contemporary homes, we revisited the experiment with new specifications to break away from the hegemonic gaze present in the software's data. For this purpose, we selected fifteen countries with distinct realities, featuring prominent architectural studios from each. By adding other elements of composition, such as space typologies or materials, we created a series of images that demonstrate different possibilities of envisioning a residence from its exterior and interior in each region.
This allowed us to incorporate different visions and preferences into the machine-learning process, making the results more personalized. We believe that these new tools can be very useful for imagining, visualizing, and designing spaces of the near future, and this project is our humble contribution to that.
In the vast quantity of images created, it is possible to glimpse how architecture can be diverse and present different solutions for the same purpose: dwelling. Vernacular traditions, further use of materials, comfort techniques employed, interiors, and even the arrangement of human scales. In the set of images, all the signs brought by artificial intelligence configure a tool that helps represent the culture of each context and envision different spatial practices. Furthermore, it is possible to notice patterns that repeat in some images, leading to brief conclusions: ancestral techniques translated into contemporaneity, the use of biophilic design in practically all projects, and inefficient spatial solutions or those without a clear function that an architect would never realize.
In times when an image no longer necessarily corresponds to reality - and can be used to distort it - it is not enough to rely solely on AI or accept everything seen as valid. When dealing with a tool that influences how we perceive the world, seeking a critical eye to create or understand each image is crucial. While artificial intelligence presents valuable resources in the design process, the human touch will still define the potential behind everything being represented and imagined.
Text of the article by our dear friend Victor Delaqua, without whom this article would have never happened.